Cudadma: Overview and Code Examples

What is cudaDMA?
An API for efficiently copying data from global to shared memory
Allows experts to implement optimized tuned code, accessible through a
simple API
Decouples size & shape of thread block from size & shape of data
Works with or without warp specialization
Why did we write this?
To research productive high-performance GPU programming techniques
Where can I find it?
http://code.google.com/p/cudadma/